To calculate the relative risk (RR) between the vaccinated and unvaccinated populations, we used the age-standardized rates indicated in the excel files provided by the UK ONS. ⁵ Their 95% confidence intervals (CI) were calculated according to the following formula ^{8 , 9} : CI 95 ( RR ) = e [ ln ( RR ) ± 1.96 ∗ SE ln ( RR ) ] , where “ln (RR)” is the natural logarithm of the Relative Risk and “SE _ln(RR)” is the standard error of the natural logarithm of the RR.

The SE _ln(RR) was calculated for each vaccination status of each of the age groups according to the formula: SEln ( RR ) = ( V . Pop . − Stand . D . Exp . V . Pop . ∗ Stand . D . Exp . ) + ( Un . Pop . − Stand . D . Exp . Un . Pop . ∗ Stand . D . Exp . ) , (1)

Where, for each year, “V.Pop.” represents the vaccinated population, “Un.Pop.” represents the unvaccinated population, and “Stand.D.Exp.” represents the expected standardized deaths, that is, the deaths that would occur by applying the Age-standardised mortality rates per 100,000 person-years (Note 1, spreadsheet “Note” UK ONS web-based platform ⁶ ) to the real population, calculated according to the formula: Expected Standardized Deaths = Age − standardized rate ∗ Population of each vaccine status 100.000 .

The choice to use the “Expected Standardized Deaths” is justified by the fact that the calculated RR expresses the ratio between two standardized rates based on the European population. The P value was calculated according to Altman and Bland ¹⁰ : p = exp ( − 0.717 × z − 0.416 × z 2 ) where z = ln ( RR ) SE and SE is the (1).

To calculate SMR indirect standardization method was applied according to Naing (2000). ¹¹ Before to use the simple regression model, all the assumption of the model was verified: scatter plot was created to verify the linear relationship between variables, Shapiro-Wilk normality test was used to verify the residual distributions and Breusch-Pagan test was used to verify homoscedasticity of the variance of errors. To calculate the intersection of the regression line with the reference line of the unvaccinated for each age group we indicated the observation months with a progressive number and solved the equation of the regression line for x and assigning y=1. The x value obtained was compared with the numbers assigned of the observation months so that we could identify a specific moment in the observation period.

Data was processed using R studio (version 2023.09.0).

The UK ONS dataset we investigated is the latest version available. ⁵ It is based on the population in Census 2021, linking Census deidentified records to National Health Service (NHS) numbers. People with no NHS number or multiple entries are not included.

The individuals were then linked via NHS number to vaccination data from the National Immunisation Management Service (NIMS) and ONS death registrations. The population was restricted to people in England, alive on 1 April 2021. Overall, ONS dataset population (51,786,812 people) covers 91.6% of the England population on Census Day 2021. The excluded population therefore amount to almost 4,600,000 people. Furthermore, 103,142 were excluded due to erroneous or inconsistent vaccination data, so the overall excluded population amounts to almost 4,700,000 people.

Finally, of the 1,149,784 deaths that occurred in England between 1 April 2021 and 31 May 2023, 90.6% (1,041,524) could be linked to individuals in the 2021 Census.

The stacked graphs showed a dynamic distribution over all observation period whose percentages and absolute frequencies are reported in Supplementary Material (Tables S1-S7 and Figures S1A/B–S7A/B) (Extended data). ³⁹

All-causes mortality rate ratios RR according to vaccination status are shown in Supplementary, Tables S8-S14(Extended data) ³⁹ and Figures S8-S14(Extended data). ³⁹ Similarly, RRs for non-COVID19 related deaths are shown in Tables S15-S21(Extended data) ³⁹ and Figures S15-S21(Extended data). ³⁹ Main results for both mortality causes are summarized below.

Deaths from all causes:

In all age groups, those vaccinated with the first dose at least 21 days ago have a significantly higher risk of death from all causes than those not vaccinated in almost all months of the entire period, except for the 18-39 age group in which the RRs are significatively higher than 1 in half the months. The average RR values in all age groups are between 1.7 and 2.3, except for the 18-39 years group where the average is 1.5. In the age groups 18-39, 60-69, and older RRs present initial peaks higher than 3, up to a maximum of 5.5 in the 70-79 age group.

As regards those vaccinated with 2 doses at least 21 days ago, in the age groups starting from 60 years, the risk of death from all causes in the initial months is much lower than the unvaccinated, with a tendency to increase. Since around a third of the entire period (between October and December 2021) RRs significantly exceeds the reference value, remaining higher in almost all the remaining months, although not significatively in the last few months for 60-69 and 90+ years age groups.

The RRs for those vaccinated with three doses at least 21 days ago, for the age groups 60-69 years and older, present a growth trend which, starting from values much lower than one, reaches and exceeds the reference value: •

for the 60-69 years age group, the risk of death significantly exceeds that of the unvaccinated in the months of November and December 2022 and remains in the following months not significantly different from that of the unvaccinated;

Deaths non correlated to COVID19

The rate ratios (RR) from non-Covid causes follow the trends already seen for deaths from all causes, with slightly higher values. Therefore, the above considerations can be repeated.

Age groups

The results of all regression models performed for the all-causes deaths and non-COVID19 deaths variable and for each age group are summarized in Table 1. Furthermore, in table 1 are indicated the intersection of the regression line with the reference line of the unvaccinated also in the graphs where is not possible to visualize the intersection.

The regression model for 18-39 age group showed coefficient of determination R ²=0.601 and a p-value <0.0001 for the All-causes death variable. For the non-COVID19 deaths regression model showed a R ²=0.500 and a p-value=0.0005 ( Figure 1). Both variables showed the intersection of the regression line in January 2023.

The regression model for 40-49 age group showed coefficient of determination R ²=0.712 and a p-value <0.0001 for the All-causes death variable. For the non-COVID19 deaths regression model showed a R ²=0.463 and a p-value=0.0003 ( Figure 2). The All-causes death variables showed the intersection of the regression line in September 2023 while the non-COVID19 deaths showed the intersection of the regression line in April 2023.

The regression model for 50-59 age group showed coefficient of determination R ²=0.734 and a p-value <0.0001 for the All-causes death variable. For the non-COVID19 deaths regression model showed a R ²=0.609 and a p-value<0.0001 ( Figure 3). The All-causes death variables showed the intersection of the regression line in July 2023 while the non-COVID19 deaths showed the intersection of the regression line in January 2025.

The regression model for 60-69 age group showed coefficient of determination R ²=0.847 and a p-value <0.0001 for the All-causes death variable. For the non-COVID19 deaths regression model showed a R ²=0.745 and a p-value<0.0001 ( Figure 4). The All-causes death variables showed the intersection of the regression line in February 2024 while the non-COVID19 deaths showed the intersection of the regression line in May 2024.

The regression model for 70-79 age group showed coefficient of determination R ²=0.860 and a p-value <0.0001 for the All-causes death variable. For the non-COVID19 deaths regression model showed a R ²=0.764 and a p-value<0.0001 ( Figure 5). The All-causes death variables showed the intersection of the regression line in January 2024 while the non-COVID19 deaths showed the intersection of the regression line in May 2024.

The regression model for 80-89 age group showed coefficient of determination R ²=0.784 and a p-value <0.0001 for the All-causes death variable. For the non-COVID19 deaths regression model showed a R ²=0.706 and a p-value<0.0001 ( Figure 6). The All-causes death variables showed the intersection of the regression line in April 2023 while the non-COVID19 deaths showed the intersection of the regression line in January 2023.

The regression model for 90+ age group showed coefficient of determination R ²=0.695 and a p-value <0.0001 for the All-causes death variable. For the non-COVID19 deaths regression model showed a R ²=0.705 and a p-value<0.0001 ( Figure 7). The All-causes death variables showed the intersection of the regression line in September 2022 while the non-COVID19 deaths showed the intersection of the regression line in May 2022.

One hypothesis might be an underestimation of the unvaccinated population in the ONS dataset. We have seen that the population included in this dataset does not cover the entire population of England recorded in the 2021 census. The population left out of the dataset is around 4,700,000 people. There would be a selection bias with systematic effects throughout the period if this proportion of the total population were not equally distributed between vaccinated and unvaccinated, and if it included a greater proportion of unvaccinated. We can see from the last two ONS reports that, although they cover a larger population than previous reports, the criteria for inclusion (and therefore exclusion) are the same as those based on the previous census. They are therefore subject to the same limitations. The main one is that the excluded population is not randomly selected, and therefore the population covered by the dataset is not representative of the general population. The key to inclusion is having a National Health Service (NHS) number. The probability of a vaccinated person not having an NHS number is virtually zero, because without a number you cannot be vaccinated. However, this is not the case for the unvaccinated, some of whom may not have been registered with a General Practitioner (GP) and therefore do not have an NHS number. As deaths among the unvaccinated are certified in the same way as those among the vaccinated, they cannot escape ONS registration. This would result in a relative overestimation of mortality rates for the unvaccinated and consequently an underestimation of RRs and SMRs for the vaccinated.

Another possible contributory hypotheses is the so called healthy-vaccinee bias. ^{16 – 19} In 2021 a lower mortality among the vaccinated can also be explained partly by the healthy-adherer effect, or the healthy-vaccinee bias in the vaccination field. This effect is much more powerful than commonly thought: in fact, voluntary adherence to a treatment can be associated with a nearly halved mortality, ^{20 – 22} and even with a mortality reduction of 2.5 to 3 or more times ^{16 , 18 , 23 – 25} compared to the mortality of those who do not adhere. This effect is independent of the type of treatment to which one adheres voluntarily, being also found in randomized controlled trials in the placebo adherers (compared with placebo non-adherers). This effect may have several explanations. In the short time, individuals contingently ill tend to postpone vaccination. ^{25 , 26} In addition, doctors can renounce to vaccinate people considered close to death, whose subsequent death burdens the unvaccinated cohort disproportionately.

However, the healthy-adherer effect can be detected over many years, ^{16 , 18 , 20 , 23} because subjects adhering to preventive treatments are usually at the same time more likely to engage in healthy lifestyles than patients not adhering. ^{17 , 21} A healthy lifestyle includes diet, exercise, lower tobacco and alcohol consumption, less risky behaviors, ²⁷ and the search for better health care. These features—difficult to capture in administrative databases—are associated with morbidity and mortality outcomes in observational studies. Moreover, the trust in the intervention to which one adheres can exert a beneficial placebo effect. The healthy-adherer bias is more difficult to correct than the opposite effect of confounding by indication (subjects in worse health conditions are vaccinated first), which is relatively easier to correct, provided it is known e.g. the number of comorbidities, or the Charlson comorbidity index of the groups to be compared. ²⁸ Unfortunately, the UK ONS public data do not include any information about comorbidities. It is likely that the healthy-vaccinee bias effect will continue to operate to varying degrees in 2022 and 2023, albeit to a diminishing extent during periods in which vaccination mandates have been in force.

Moreover, it is plausible that the increase in the number of the vaccinated has diluted the opposite effect of confounding by indication. ²⁸

A commonly used argument is the better known confounding by indication effect ²⁸ : it is likely that fragile subjects with multiple diseases have been vaccinated as a priority, followed by the others. However, as the vaccination campaigns proceed, the composition of the vaccinated and unvaccinated populations should result less unbalanced with respect to the pre-existing state of health. The ONS declare that “Changes in non-COVID-19 mortality by vaccination status are largely driven by the changing composition of the vaccination status groups. This is because of the priority given to clinically extremely vulnerable people or with underlying health conditions, and differences in timing of vaccination among eligible people". ³ However, a priority was also given to the healthier population of health workers. Moreover, the most fragile part of the population prioritized for the vaccination is a smaller portion (especially in the younger ages), and the composition of each age group progressively tends to be similar to that of the unvaccinated, in terms of general health conditions. Therefore, a decreasing trend would be expected, both because of the decreasing weight of the fragile fraction compared to the overall group and because of the harvesting effect, described below.

The healthy-vaccinee/un-healthy-un-vaccinee bias might be further supported by the web page by the Office for National Statistics, ³¹ showing that unvaccinated have higher tendency: to live in more deprived areas, urban areas, or social rented housing, to be not born in the UK or do not have English as a main language, to have never worked or to be long-term unemployed, to be more limited by a disability, and to be male (more men die than women). ³¹

The RRs of the first doses generally show high initial mortality peaks, possibly linked to the priority given to the fragile subjects. The early death of the most fragile causes that those who move on to the second dose are healthier overall. It is unfortunate that the last two ONS datasets do not provide data for the first three months of 2021, corresponding to the start of the vaccination programme. In fact, since vaccinations began with the older classes, in April 2021 the latter had not only completed the vaccination with the first dose, but had already started the second. Therefore our hypothesis can be confirmed only in the age groups 18-39 and 40-49, where one can clearly perceive the initial mortality peak of the first doses, probably already decreasing, followed by the initial mortality peak of the second doses, starting from lower values and with a lower maximum compared to that of the first doses.

As regards the third doses, the initial mortality peak disappears in all age groups, suggesting that many of the most “fragile” people have already died, and that a ‘healthy-vaccinee effect’ might partly explain the initial very low RR values.

Examining the SMR graphs, we note that, apart from the elderly, the points relating to all-cause deaths and non-COVID-19 deaths tend to overlap over time. This may indicate that the impact of COVID-19-related deaths vanishes, and that the risk of all-cause death and of deaths not related to COVID-19 is nearly the same. In fact, in the ONS dataset, approaching the end of the observation period, the COVID-19-deaths for the different vaccination statuses show an ever-increasing number of values indicated with <3, what prevents one from calculating both the RRs and the SMRs. This justifies the choice not to take into consideration COVID-19 deaths, but only all-cause and non-COVID-19 deaths.

Hence, the insistent push towards further vaccinations seems hardly motivated.

Again, examining the SMRs, the fact that initially the regression line of non-COVID-19 deaths is above that of all-cause deaths might indicate that the vaccine initially has a protective effect on COVID-19 deaths, thus lowering the risk of all-cause deaths, that include those COVID-19 related, among vaccinated people. The fact that they subsequently converge may indicate either that the vaccine gradually loses its protective effectiveness, or that the risk of COVID-19 deaths decreases due to the increasingly lower lethality of the new variants, or that the two causes act together.

The SMR graphs allow one to make a further consideration: in the 80-89 age group the convergence of the lines of all-cause deaths is much attenuated, and in the 90+ age group the lines are almost parallel, as one can see from the regression coefficients that differ less and less. This might indicate that COVID-19 still represents a risk for the elderly and that the vaccine therefore protects them by reducing the risk of all-cause deaths. Yet, it might also be due to the fact that the lower lethality of the new variants is offset by the fact that the vaccinated people get infected more than unvaccinated ^{29 – 36} and that by taking additional doses they are temporarily protected from the risk of dying from COVID-19.

Last but not least: why are the SMRs of non-COVID-19 related deaths increasing? Why should the risk of those vaccinated with any dose increase compared to that of the unvaccinated? Apart from the risk of immediate adverse reactions/events, the doubt naturally arises that the vaccine , might cause damage to the immune system, exposing the vaccinated to a greater risk of death from pathologies non-COVID-19 related ^{37 , 38}